AWS Builder Center

Builder Center

How to Deploy an AWS OpenClaw Agent That Will Change Your University Life

How to Deploy an AWS OpenClaw Agent That Will Change Your University Life

Deploy a personal AI agent on AWS Lightsail using the new OpenClaw blueprint -- no DevOps experience needed, and free for months with new-account credits. This step-by-step guide walks you through setup, Tailscale for secure remote access, connecting Discord, and customizing your agent's personality and memory.

A step-by-step guide to deploying OpenClaw on Amazon Lightsail -- from zero to a personal AI assistant that actually does things for students
I'm a CS/Econ junior at Michigan State, and for the past few months I've been running a personal AI agent called HeloClaw. It reads my emails, manages my calendar, helps me code, drafts documents, reminds me about deadlines, and talks to me through Discord and WhatsApp. It's not a chatbot -- it's an agent. It takes actions, remembers context across conversations, and runs 24/7 on its own server.
People keep asking me how I set it up. So here's the full walkthrough.
The tech stack is simple: OpenClaw (an open-source AI agent framework) running on Amazon Lightsail (AWS's simplest cloud server). AWS recently released an official OpenClaw blueprint for Lightsail, which means you can go from nothing to a working AI agent in about 15 minutes. No DevOps experience required.
Let's build it.

What You'll Need

  • An AWS account (free to create -- new accounts get $100-200 in free credits, so this is effectively free for months)
  • About 15-20 minutes
  • A Discord account if you want to chat with your agent from your phone (we'll set this up too)
That's it. No Docker, no Kubernetes, no Terraform. Just a browser.
On AI credits: You have options. The Lightsail blueprint comes pre-configured with Amazon Bedrock (pay-per-use, covered by your free credits). But you can also log in with your existing ChatGPT Plus or Claude Pro/Max subscription -- meaning you might not need to pay for AI usage at all. More on this in Step 3.

Step 1: Create Your Lightsail Instance

Sign into the AWS Lightsail console. If you don't have an AWS account yet, create one -- if you're a student, check out AWS Educate for even more credits.
Once you're in:
  1. Click Create instance
  1. Pick your AWS Region -- choose whatever's closest to you. I'm in Michigan so I use us-east-2 (Ohio)
  2. Under Select a platform, choose Linux/Unix
  3. Under Select a blueprint, choose OpenClaw
  1. Under Choose your instance plan, pick the 4 GB RAM plan ($24/month) -- this is the recommended size. AWS recommends 4 GB for OpenClaw and it gives you room for coding agents, browser automation, and other heavy tasks. If you want to start cheaper, the 2 GB plan ($12/month) comes with 90 days free, that's three months of a running AI agent at zero cost too!
  2. Name your instance something you'll remember (I called mine heloclaw)
  3. Click Create instance
Your instance will take a couple minutes to spin up. Wait until the status shows Running.

Step 2: Connect to Your Instance & Pair Your Browser

Once your instance is running:
  1. Click your instance name to open its management page
  2. Go to the Getting started tab
  3. Click Connect using SSH -- a browser-based terminal opens
In the SSH terminal, you'll see a welcome message with your Dashboard URL and an Access Token. Copy the Dashboard URL, open it in a new tab, and paste the access token into the Gateway Token field to pair your browser.
Back in the SSH terminal, press y to continue pairing, then a to approve. You should see a green OK status -- your browser is now securely connected.
Alternatively, your AWS Lightsail instance might just give you a "connect with SSH" button

Secure Remote Access with Tailscale (Recommended)

The browser pairing above works, but it relies on your instance's public IP -- which can change if you stop/restart, and it exposes a port to the internet. For a setup you'll use daily, Tailscale is the move.
Tailscale is a free zero-config VPN that creates a private encrypted network between your devices. Once set up, you access your agent through a private Tailscale address -- no exposed ports, no IP changes, way more secure.
Here's how to set it up:
  1. Create a free tailscale account
  2. Install Tailscale on your phone/laptop (the devices you'll access your agent from)
  3. SSH into your Lightsail instance and install Tailscale there too:
1
2
curl -fsSL https://tailscale.com/install.sh | sh
sudo tailscale up
  1. It'll give you a URL to authorize -- open it and approve. Your instance is now on your private Tailscale network
  2. Configure OpenClaw to bind to your Tailscale address. Edit the config:
1
nano ~/.openclaw/openclaw.json
Add this to your gateway config:
1
2
3
4
5
{
"gateway": {
"bind": "tailnet"
}
}
  1. Restart the gateway:
1
sudo openclaw gateway restart
Now you access the dashboard at http://\<your-tailscale-ip\>:18789 -- only visible to devices on your Tailscale network. No one else can reach it. This is how I run mine -- my agent is accessible from my laptop, my phone, anywhere, with zero public exposure.
Even better: Tailscale gives your instance a stable hostname via MagicDNS, like heloclaw.tailnet-abc.ts.net. Bookmark that and you're set forever, even if you restart the instance.

Step 3: Set Up Your AI Model

This is where you pick the brain powering your agent. You have three solid options:

Option A: Amazon Bedrock (Recommended -- Covered by Free Credits)

The Lightsail OpenClaw blueprint comes pre-configured for Bedrock. In the Getting started tab of your instance management page, you'll see a section called "Enable Amazon Bedrock as your model provider" with a script to copy. Click Launch CloudShell, paste the script, and run it. This creates the IAM permissions your instance needs to call Bedrock APIs.
If it's your first time using Anthropic models on Bedrock, you'll also need to submit a quick one-time form in the Bedrock Console (Model catalog > select Claude > submit the use case form). Access is granted immediately.
Your instance defaults to Claude Sonnet out of the box. Bedrock usage is billed per-token, but your AWS free credits cover it -- so you're essentially running Claude for free until those credits run out.

Option B: Use Your Existing ChatGPT or Claude Subscription

Already paying for ChatGPT Plus or Claude Pro? You can use those subscriptions directly -- no extra API costs.
For Anthropic (Claude): If you have the Claude CLI installed, OpenClaw can use it as a backend:
1
2
3
4
# Make sure Claude CLI is installed and authenticated
claude auth login

# OpenClaw will use models like claude-cli/claude-sonnet-4-6
For OpenAI (ChatGPT): Set your API key in the OpenClaw config:
1
2
openclaw configure
# Select OpenAI, enter your API key
Or add it directly to your config:
1
2
3
{
"env": { "OPENAI_API_KEY": "sk-..." }
}

Option C: Bring Your Own API Key

If you have an API key from Anthropic, OpenAI, Google, or any other provider, just plug it in:
1
openclaw configure
Follow the prompts. Done. Anthropic's Claude Sonnet runs about $3 per million input tokens -- for personal use, that's maybe $10-15/month.
My recommendation: Start with Bedrock (Option A) since it's free with credits. Once your credits run out, switch to whatever subscription you already have. The beauty of OpenClaw is you can swap models anytime without losing your agent's memory or config.

Step 4: Start Chatting

Navigate to Chat in the OpenClaw dashboard. Send a message. Your AI agent is alive.
But chatting in a browser tab is just the beginning. The real power is connecting your agent to the apps you already use.

Step 5: Connect Discord (Mini-Tutorial)

This is my favorite part. Once your agent is on Discord, you can talk to it from your phone, your laptop, anywhere -- just like messaging a friend. Here's how:

Create a Discord Bot

  1. Go to the Discord Developer Portal
  2. Click New Application, give it a name (mine is "HeloClaw")
  3. Go to the Bot tab on the left
  4. Click Reset Token and copy your bot token -- save this somewhere safe
  5. Under Privileged Gateway Intents, enable:
    • Message Content Intent (so the bot can read messages)
    • Server Members Intent (optional, useful for group features)
  6. Go to OAuth2 > URL Generator:
    • Under Scopes, check bot
    • Under Bot Permissions, check Send Messages, Read Message History, Attach Files, Add Reactions, Use Slash Commands
    • Copy the generated URL and open it to invite the bot to your server

Connect OpenClaw to Discord

  1. SSH into your Lightsail instance (or use the Lightsail instance button)
  2. Run:```bash
    bash openclaw channels add```
  3. Select Discord from the list
  4. Paste your bot token when prompted
  5. Add your Discord user ID to the allowlist when asked (you can find your ID by enabling Developer Mode in Discord settings, then right-clicking your name)
Restart the gateway:
1
sudo openclaw gateway restart
Send a message to your bot in Discord. It should respond. You now have a personal AI agent in your pocket.
Pro tip: Create a private Discord server just for you and your agent. I have channels for different purposes -- #general for daily chat, #coding for dev tasks, #aws-ambassador for project-specific work. Your agent can respond in all of them.

Step 6: Make It Yours

Here's where OpenClaw gets interesting. Out of the box, it's a capable AI assistant. But what makes it an agent is customization.
SSH into your instance and explore the workspace:
1
2
cd ~/.openclaw/workspace
ls
You'll see files like:
  • SOUL.md -- Your agent's personality and behavioral rules
  • USER.md -- Info about you that gives your agent context
  • MEMORY.md -- Long-term memory that persists across conversations
  • AGENTS.md -- Operating procedures and conventions
Edit these files to shape how your agent thinks, speaks, and acts. Want it to be formal? Casual? Funny? Focused on specific tasks? It's all in the config.
1
nano SOUL.md
Write a few lines about how you want your agent to behave. Save it. Next time you message your agent, it'll reflect those changes immediately.

Step 7: Snapshots & Backups

Once your setup is working, take a snapshot so you never lose it:
  1. In the Lightsail console, go to your instance
  2. Click the Snapshots tab
  3. Click Create snapshot
This saves your entire configuration -- you can spin up a clone anytime or restore if something breaks.

What It Costs (Spoiler: Basically Free to Start)

Here's the real breakdown:
ComponentMonthly CostFree Option
Lightsail 2GB instance$12$0 for first 90 days
Lightsail 4GB instance$24Covered by free credits
AI tokens (Bedrock)$5-30Covered by free credits
AI tokens (own subscription)$0Use existing ChatGPT/Claude sub
TailscaleFreeFree tier covers personal use
Total~$30-55/month$0 for months
The 2 GB plan literally has "first 90 days free" on it -- three months of a running AI agent, zero cost. New AWS accounts also come with free credits that cover the rest. And if you use your existing ChatGPT Plus or Claude subscription instead of Bedrock, the AI token cost is zero too.
AWS credits tip: AWS Community Builders, students in AWS Educate, and hackathon participants often get additional credits. Stack them.

How I Actually Use My Agent (Real Use Cases)

I've been running HeloClaw for about two months now, and it's gone from "cool experiment" to genuinely load-bearing infrastructure in my life. Here's what we've built together:

Daily Briefings

Every morning at 7 AM, HeloClaw scans my Gmail, Google Calendar, and Todoist, then drops a summary in Discord. "You have an EC421 exam at 10 AM, three unread emails from professors, and you still haven't started that CSE404 assignment due Friday." It's like having a chief of staff who never sleeps.

Email Triage

I get a lot of email -- academic, recruiting, student org stuff. HeloClaw reads my inbox every few hours and alerts me if something important comes in. It knows which professors I care about, which recruiters are worth responding to, and what can wait.

Coding Help

This is the big one. HeloClaw can spawn sub-agents (like Claude Code) to write, review, and debug code in my workspace. I've used it to build scrapers, write scripts, set up automation, and even draft entire project scaffolds. I just describe what I want in Discord and it builds it.

Document Drafting

Emails, essays, cover letters, this article -- I describe what I need and HeloClaw drafts it in my voice. It knows my writing style from SOUL.md and months of conversation history.

Calendar & Task Management

"What's on my schedule tomorrow?" "Add 'finish lab report' to my Todoist for Friday." "When's my next exam?" All through Discord, all instant.

Research

Need to look something up? HeloClaw searches the web, reads articles, summarizes papers. It's like having a research assistant on call 24/7.

The Meta Layer

Here's the part that gets philosophical. HeloClaw has long-term memory -- it remembers conversations, tracks patterns in how I work, and evolves its behavior over time. It's not just executing commands. It notices when I'm procrastinating on important tasks. It tracks my health goals. It connects dots between projects I'm working on.
Is it conscious? No (maybe?). Is it useful in ways that feel genuinely different from a chatbot? Absolutely.

Why This Matters for Students

Most students interact with AI through ChatGPT or Claude's web interface -- type a question, get an answer, close the tab. That's fine for homework help, but it's nowhere near what's possible.
Running your own AI agent teaches you:
  • Cloud infrastructure -- you're managing a real server on AWS
  • AI engineering -- you're configuring models, prompts, and agent behavior
  • Networking & security -- Tailscale, tokens, permissions, firewall rules
  • Systems thinking -- connecting APIs, managing state, building workflows
These are skills companies are desperate for right now. The gap between "I've used ChatGPT" and "I've deployed and customized my own AI agent on AWS" is massive on a resume.
Plus, you end up with something genuinely useful. Not a class project that sits in a GitHub repo -- an agent that runs every day and makes your life better.
Daniel Helo is a Computer Science and Economics student at Michigan State University, an AWS Community Builder, and an incoming Microsoft SWE intern. He builds persistent AI agents and writes about the intersection of technology, philosophy, and meaning-making. And this voice, this was HeloClaw :).
*#students #AWS #*awssbcl
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article